Instructions to use Cyclone-Labs/Twilight-Embrace-31B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Cyclone-Labs/Twilight-Embrace-31B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="Cyclone-Labs/Twilight-Embrace-31B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("Cyclone-Labs/Twilight-Embrace-31B")
model = AutoModelForMultimodalLM.from_pretrained("Cyclone-Labs/Twilight-Embrace-31B", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Cyclone-Labs/Twilight-Embrace-31B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Cyclone-Labs/Twilight-Embrace-31B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cyclone-Labs/Twilight-Embrace-31B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Cyclone-Labs/Twilight-Embrace-31B

SGLang

How to use Cyclone-Labs/Twilight-Embrace-31B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Cyclone-Labs/Twilight-Embrace-31B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cyclone-Labs/Twilight-Embrace-31B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Cyclone-Labs/Twilight-Embrace-31B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cyclone-Labs/Twilight-Embrace-31B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use Cyclone-Labs/Twilight-Embrace-31B with Docker Model Runner:
```
docker model run hf.co/Cyclone-Labs/Twilight-Embrace-31B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Overview

Twilight-Embrace-31B was created by combining gemma-4-31B-it, Eon-Blossom-V2-31B, Love-Aligned-31B, and Dark-Scarlett-v2.0-31B using a custom merge method.

Merge Configuration

base_model: google/gemma-4-31B-it
models:
  - model: Cyclone-Labs/Eon-Blossom-V2-31B
    parameters:
      weight:
        - filter: self_attn.q_proj.weight
          value: [0.72, 0.72, 0.71, 0.7, 0.68, 0.68, 0.7, 0.71]
        - filter: self_attn.k_proj.weight
          value: [0.73, 0.72, 0.71, 0.7, 0.69, 0.69, 0.7, 0.72]
        - filter: self_attn.v_proj.weight
          value: [0.23, 0.22, 0.21, 0.2, 0.19, 0.18, 0.18, 0.19]
        - filter: self_attn.o_proj.weight
          value: [0.54, 0.52, 0.5, 0.48, 0.46, 0.45, 0.47, 0.5]
        - filter: mlp.gate_proj.weight
          value: [0.2, 0.19, 0.18, 0.17, 0.16, 0.16, 0.18, 0.2]
        - filter: mlp.up_proj.weight
          value: [0.2, 0.19, 0.18, 0.17, 0.16, 0.16, 0.18, 0.2]
        - filter: mlp.down_proj.weight
          value: [0.54, 0.52, 0.5, 0.48, 0.46, 0.45, 0.47, 0.5]
        - filter: layer_scalar
          value: [0.8, 0.8, 0.79, 0.78, 0.78, 0.78, 0.79, 0.8]
        - filter: model.language_model.layers.
          value: [0.75, 0.75, 0.74, 0.73, 0.72, 0.72, 0.73, 0.75]
        - filter: language_model.embed_tokens.weight
          value: 0.4
        - filter: language_model.norm.weight
          value: 0.8
        - value: 0
  - model: UnstableLlama/Love-Aligned-31B
    parameters:
      weight:
        - filter: self_attn.q_proj.weight
          value: [0.23, 0.23, 0.24, 0.25, 0.27, 0.27, 0.25, 0.24]
        - filter: self_attn.k_proj.weight
          value: [0.23, 0.24, 0.25, 0.26, 0.27, 0.27, 0.26, 0.24]
        - filter: self_attn.v_proj.weight
          value: [0.22, 0.23, 0.24, 0.25, 0.26, 0.27, 0.27, 0.26]
        - filter: self_attn.o_proj.weight
          value: [0.41, 0.43, 0.45, 0.47, 0.49, 0.5, 0.48, 0.45]
        - filter: mlp.gate_proj.weight
          value: [0.25, 0.26, 0.27, 0.28, 0.29, 0.29, 0.27, 0.25]
        - filter: mlp.up_proj.weight
          value: [0.25, 0.26, 0.27, 0.28, 0.29, 0.29, 0.27, 0.25]
        - filter: mlp.down_proj.weight
          value: [0.41, 0.43, 0.45, 0.47, 0.49, 0.5, 0.48, 0.45]
        - filter: layer_scalar
          value: [0.17, 0.17, 0.18, 0.19, 0.19, 0.19, 0.18, 0.17]
        - filter: model.language_model.layers.
          value: [0.2, 0.2, 0.21, 0.22, 0.23, 0.23, 0.22, 0.2]
        - filter: language_model.embed_tokens.weight
          value: 0.3
        - filter: language_model.norm.weight
          value: 0.17
        - value: 0
  - model: ReadyArt/Dark-Scarlett-v2.0-31B
    parameters:
      weight:
        - filter: self_attn.q_proj.weight
          value: 0.05
        - filter: self_attn.k_proj.weight
          value: 0.04
        - filter: self_attn.v_proj.weight
          value: 0.55
        - filter: self_attn.o_proj.weight
          value: 0.05
        - filter: mlp.gate_proj.weight
          value: 0.55
        - filter: mlp.up_proj.weight
          value: 0.55
        - filter: mlp.down_proj.weight
          value: 0.05
        - filter: layer_scalar
          value: 0.03
        - filter: model.language_model.layers.
          value: 0.05
        - filter: language_model.embed_tokens.weight
          value: 0.3
        - filter: language_model.norm.weight
          value: 0.03
        - value: 0
merge_method: arcus
chat_template: auto
parameters:
  gain: 0.95
  inheritance: 1.1
  declone: 0.15
  anchor: 0.35
  containment: 1.15
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: base

Notes

This was an interesting result. Love-Aligned-31B added a unique flavor to the merge. It is adversarial and will disagree with you in roleplay, but it seems to always seek a restorative ending in long-form story writing tests. This model can still be dark and realistic.

Intended Use

Roleplay

Character-driven interaction, personas, dialogue, emotional scenes, and long-form roleplay.

Creative Writing

Fiction, dialogue, atmosphere, descriptive writing, stylistic drafting, and imaginative prose.

Storytelling

Long-form narratives, worldbuilding, continuity, evolving characters, and multi-character plots.

Interactive Fiction

Branching narratives, scenario play, character interaction, and continuously evolving stories.

Downloads last month: 215

Safetensors

Model size

31B params

Tensor type

BF16

Model tree for Cyclone-Labs/Twilight-Embrace-31B

Cyclone-Labs/Eon-Blossom-V2-31B

ReadyArt/Dark-Scarlett-v2.0-31B

UnstableLlama/Love-Aligned-31B

google/gemma-4-31B-it

Merge model

this model

Quantizations

4 models

Cyclone-Labs
/

Twilight-Embrace-31B

Twilight-Embrace-31B